CN110517259A - A kind of detection method, device, equipment and the medium of product surface state - Google Patents
A kind of detection method, device, equipment and the medium of product surface state Download PDFInfo
- Publication number
- CN110517259A CN110517259A CN201910816859.2A CN201910816859A CN110517259A CN 110517259 A CN110517259 A CN 110517259A CN 201910816859 A CN201910816859 A CN 201910816859A CN 110517259 A CN110517259 A CN 110517259A
- Authority
- CN
- China
- Prior art keywords
- result
- detection
- surface state
- photo
- defects detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/0008—Industrial image inspection checking presence/absence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Abstract
This application discloses detection method, device, equipment and the media of a kind of product surface state, are related to field of cloud calculation, artificial intelligence field.Specific implementation are as follows: obtain target product to be detected under at least two shooting angle, multiple the surface state photos shot at least two intensities of illumination；Each surface state photo is separately input into corresponding detection model, and according to the output of each detection model as a result, obtaining defects detection result corresponding with each surface state photo；Wherein, the detection model use is obtained in the case where setting shooting angle for the sample image training that setting intensity of illumination is shot；According to defects detection corresponding with each surface state photo as a result, determining the testing result of the target product.The technical solution of the embodiment of the present application may be implemented to carry out automatic detection to product according to the surface state photo of product, improve detection efficiency and detection accuracy.
Description
Technical field
The invention relates to computer technology more particularly to artificial intelligence technologys, and in particular to a kind of product surface
Detection method, device, equipment and the medium of state.
Background technique
It is manufactured in scene in traditional notebook computer, the surface state detection of product is manufacturer's control shipment matter
Amount, the important link for maintaining the relations of production.This traditional notebook manufacturer, by the state-detection to notebook surface, with judgement
Notebook whether there is flaw and defect, perform corresponding processing according to testing result to notebook, to realize quality control.
Common deficiency type has water spot, lousiness, black print etc..
In the prior art, the quality inspection based on product surface state is mostly pure artificial visual quality inspection or semi-automatic optical instrument
The artificial quality inspection of auxiliary.In the case where pure artificial visual quality inspection, the business expert of needs is checked, artificial after discovery defect
It records and does subsequent processing again.This method not only low efficiency, is easy erroneous judgement of failing to judge, while the industrial number that this mode generates
According to being not easy to store, count, manage and mining again recycles.Semi-automatic optical instrument auxiliary artificial quality inspection the case where
Under, feature and decision rule are all based on experience and are cured in machine, it is difficult to the development iteration of business, cause with production
The detection accuracy of the development of technique, system is lower and lower, or even is reduced to complete unusable state.In addition, feature and judgement
Rule is all solidified by third-party vendor within hardware in advance, and when upgrading not only needs to carry out production line key technological transformation, but also
It is expensive.
Summary of the invention
The embodiment of the present application provides detection method, device, equipment and the medium of a kind of product surface state, realizes root
Automatic detection is carried out to product according to the surface state photo of product, improves detection efficiency and detection accuracy.
In a first aspect, the embodiment of the present application provides a kind of detection method of product surface state, comprising:
Target product to be detected is obtained under at least two shooting angle, shoots to obtain at least two intensities of illumination
Multiple surface state photos；
Each surface state photo is separately input into corresponding detection model, and according to each detection model
Output is as a result, obtain defects detection result corresponding with each surface state photo；
Wherein, the detection model use is in the case where setting shooting angle, the sample shot for setting intensity of illumination
Image training obtains；
According to defects detection corresponding with each surface state photo as a result, determining the inspection of the target product
Survey result.
The technical solution of the embodiment of the present application, by obtaining target product to be detected under at least two shooting angle,
Multiple the surface state photos shot at least two intensities of illumination；Then each surface state photo is separately input into
In corresponding detection model, and according to the output of each detection model as a result, obtaining corresponding with each surface state photo
Defects detection result；Detection model use is in the case where setting shooting angle, for the sample image that shoots of setting intensity of illumination
Training obtains；Finally, according to defects detection corresponding with each surface state photo as a result, determining the detection of target product
As a result, can be obtained based on use in the case where setting shooting angle for the sample image training that setting intensity of illumination is shot
Detection model, obtain defects detection corresponding from each surface state photo as a result, can be by different detection models
Testing result, the testing result of a final target product is obtained, so that the testing result of target product is with reference to difference
The testing result of detection model, testing result is more accurate, improves detection accuracy, can be according to the surface state photo of product
Automatic detection is carried out to product, improves detection efficiency.
Optionally, according to defects detection corresponding with each surface state photo as a result, determining the target
The testing result of product, comprising:
According to defects detection corresponding with each surface state photo as a result, and with each detection mould
The corresponding measurement of type refers to weight, determines the testing result of the target product.
The advantages of this arrangement are as follows: by according to defects detection corresponding with each surface state photo as a result,
And measurement corresponding with each detection model refers to weight, determines the testing result of target product, it can be according to difference
The testing result of detection model and the corresponding measurement of each detection model refer to weight, obtain a final target
The testing result of product, so that the testing result of target product is with reference to the testing result and measurement of different detection models with reference to power
Weight, testing result is more accurate, improves detection accuracy.
Optionally, according to defects detection corresponding with each surface state photo as a result, and with each institute
The corresponding measurement of detection model is stated with reference to weight, determines the testing result of the target product, comprising:
Will defects detection corresponding with each surface state photo as a result, being separately input into decision model, and obtain
The testing result of the decision model output；
Wherein, the decision model learns measurement corresponding with each detection model with reference to weight in advance, described
With reference to weight, the iteration in the use process of the decision model updates for measurement.
The advantages of this arrangement are as follows: learn measurement corresponding with each detection model in advance by decision model
With reference to weight, so that final result is not only with reference to the testing result of different detection models, but also each detection mould has been combined
The detection precision of type, testing result are more accurate.
Optionally, according to defects detection corresponding with each surface state photo as a result, and with each institute
The corresponding measurement of detection model is stated with reference to weight, determines the testing result of the target product, comprising:
Obtain defects detection results set corresponding with multiple surface state photos under currently processed shooting angle；
In the defects detection results set, weight and defect are referred to according to the measurement of each defects detection result
Similarity between testing result gives a mark to each defects detection result；
According to marking as a result, the matched local testing result of the determining and currently processed shooting angle；
Local testing result under at least two shooting angle is merged into processing, obtains the target product
Testing result.
The advantages of this arrangement are as follows: weight and defects detection knot are referred to according to the measurement of each defects detection result
Similarity between fruit gives a mark to each defects detection result, according to marking as a result, determining and currently processed shooting angle
Matched part testing result, and further by the way that the local testing result under at least two shooting angle is merged place
Reason, obtains the testing result of target product, so that overlapping journey of the testing result of target product with reference to defects detection result
Degree, testing result is more accurate, and testing result is more accurate.
Optionally, according to defects detection corresponding with each surface state photo as a result, determining the target
The testing result of product, comprising:
Obtain defects detection results set corresponding with multiple surface state photos under currently processed shooting angle；
Each defects detection result in the defects detection results set is mapped in same photo；
According to the similarity of the defects detection result each in the mapping result, the determining and currently processed shooting
The local testing result of angle automatching；
Part detection under at least two shooting angle is combined and merges processing, obtains the target product
Testing result.
The advantages of this arrangement are as follows: by obtaining and multiple surface state photos pair under currently processed shooting angle
The defects detection results set answered maps to each defects detection result in defects detection results set in same photo, so
It is determining to be detected with the matched part of currently processed shooting angle afterwards according to the similarity of defects detection result each in mapping result
As a result；Processing is merged finally, the part detection under at least two shooting angle is combined, obtains the detection knot of target product
Fruit can obtain the testing result of a final target product, so that target product according to the similarity of defects detection result
Testing result with reference to the overlapping degree of defects detection result, testing result is more accurate, improves detection accuracy.
Optionally, each surface state photo is separately input into corresponding detection model, and according to each inspection
The output of model is surveyed as a result, obtaining defects detection result corresponding with each surface state photo, comprising:
It carries out currently processed surface state photo to cut figure, obtains the local state photo of multiple target sizes, it is described
Target size and the minimum defect size of the target product match；
Multiple described local state photos are separately input into and the currently processed matched inspection of surface state photo
It surveys in model, obtains defects detection result corresponding with each local state photo；
Defects detection result corresponding with each local state photo is combined, is obtained and the current place
The corresponding defects detection result of the surface state photo of reason.
The advantages of this arrangement are as follows: according to the minimum defect size of target product, surface state photo is carried out to cut figure,
Obtain the local state photo of multiple target sizes, then according to the defects detection of each local state photo as a result, obtain with
The corresponding defects detection of surface state photo is as a result, improve the accuracy of defects detection result.
Optionally, the detection model is the model generated based on Fast R-CNN algorithm；
In the detection model, convolutional layer is constructed using deformable convolution, uses focal loss function as loss
Function extracts the characteristics of image of input photo using feature pyramid network algorithm；And
Negative sample image in the sample image is used in what line difficulty sample mining algorithm excavated.
The advantages of this arrangement are as follows: detection model is the model generated based on Fast R-CNN algorithm, can be to setting
The robustness with higher such as the deformation for the sample image that intensity of illumination is shot, fuzzy, illumination variation, for classification task
With it is higher can generalization；In detection model, uses focal loss function as loss function, solve data sample class
Not unbalanced problem；The characteristics of image of input photo is extracted using feature pyramid network algorithm, improves defect recall rate；Make
The negative sample image in sample image is excavated with online difficult sample mining algorithm, improves detection model precision.
Optionally, the target product is laptop；
Multiple surface state photos that the target product is shot under at least two shooting angle, comprising: notes
This computer full face, laptop side photo and laptop turning photo.
The advantages of this arrangement are as follows: so that the real-time example of the application be suitable for it is all based on laptop image data come
The scene of testing product quality.
Second aspect, the embodiment of the present application also provides a kind of detection devices of product surface state, comprising:
Photo obtains module, for obtaining target product to be detected under at least two shooting angle, at least two
Multiple surface state photos that a intensity of illumination is shot；
As a result module is obtained, for each surface state photo to be separately input into corresponding detection model, and root
According to the output of each detection model as a result, obtaining defects detection result corresponding with each surface state photo；
Wherein, the detection model use is in the case where setting shooting angle, the sample shot for setting intensity of illumination
Image training obtains；
As a result determining module, for according to defects detection corresponding with each surface state photo as a result, really
The testing result of the fixed target product.
The technical solution of the embodiment of the present application, by obtaining target product to be detected under at least two shooting angle,
Multiple the surface state photos shot at least two intensities of illumination；Then each surface state photo is separately input into
In corresponding detection model, and according to the output of each detection model as a result, obtaining corresponding with each surface state photo
Defects detection result；Detection model use is in the case where setting shooting angle, for the sample image that shoots of setting intensity of illumination
Training obtains；Finally, according to defects detection corresponding with each surface state photo as a result, determining the detection of target product
As a result, can be obtained based on use in the case where setting shooting angle for the sample image training that setting intensity of illumination is shot
Detection model, obtain defects detection corresponding from each surface state photo as a result, can be by different detection models
Testing result, the testing result of a final target product is obtained, so that the testing result of target product is with reference to difference
The testing result of detection model, testing result is more accurate, improves detection accuracy, can be according to the surface state photo of product
Automatic detection is carried out to product, improves detection efficiency.
The third aspect, the embodiment of the present application also provides a kind of electronic equipment, comprising:
At least one processor；And
The memory being connect at least one described processor communication；Wherein,
The memory is stored with the instruction that can be executed by least one described processor, and described instruction is by described at least one
A processor executes, so that at least one described processor is able to carry out the product surface state of the application any embodiment offer
Detection method.
Fourth aspect is stored with the non-instantaneous of computer instruction and computer-readable deposits the embodiment of the present application also provides a kind of
Storage media, which is characterized in that the production that the computer instruction is used to that the computer to be made to execute the offer of the application any embodiment
The detection method of product surface state.
Other effects possessed by above-mentioned optional way are illustrated hereinafter in conjunction with specific embodiment.
Detailed description of the invention
Attached drawing does not constitute the restriction to the application for more fully understanding this programme.Wherein:
Fig. 1 a is the flow chart of one of the application first embodiment detection method of product surface state；
Fig. 1 b is one of the application first embodiment Fast R-CNN algorithm principle figure；
Fig. 1 c is the deformable convolution principle schematic diagram of one of the application first embodiment；
Fig. 2 is the flow chart of one of the application second embodiment detection method of product surface state；
Fig. 3 a is the flow chart of one of the application 3rd embodiment detection method of product surface state；
Fig. 3 b is the integrated stand composition of one of the application 3rd embodiment detection system of product surface state；
Fig. 4 is a kind of structural schematic diagram of the detection device for product surface state that the application fourth embodiment provides；
Fig. 5 is the block diagram for the electronic equipment for the detection method for realizing the product surface state of the embodiment of the present application.
Specific embodiment
It explains below in conjunction with exemplary embodiment of the attached drawing to the application, including the various of the embodiment of the present application
Details should think them only exemplary to help understanding.Therefore, those of ordinary skill in the art should recognize
It arrives, it can be with various changes and modifications are made to the embodiments described herein, without departing from the scope and spirit of the present application.Together
Sample, for clarity and conciseness, descriptions of well-known functions and structures are omitted from the following description.
First embodiment
Fig. 1 a is the flow chart of one of the application first embodiment detection method of product surface state, the present embodiment
Technical solution be suitable for according to product surface state-detection product quality the case where, this method can be by product surface state
Detection device executes, which can be realized by the mode of software and/or hardware, and generally can integrate in electronic equipment
In, specifically includes the following steps:
Step 101 obtains target product to be detected under at least two shooting angle, at least two intensities of illumination
Shoot multiple obtained surface state photos.
In the present embodiment, by automated machine equipment, the industry camera of erection and the light source design of profession, In are utilized
Under at least two shooting angle, target product to be detected is shot at least two intensities of illumination, obtains multiple surface state
Photo.Illustratively, shooting angle is front, side and the turning of target product to be detected.It is set by automated machine
For in the front of target product to be detected, shoot to obtain the full face of three target products, In for three intensities of illumination
The side of target product to be detected, for the side photo for three target products that three intensities of illumination are shot, to
The turning of the target product of detection, for the turning photo for three target products that three intensities of illumination are shot.It amounts to and claps
It takes the photograph to obtain nine surface state photos.
Optionally, target product is laptop.Target product shot under at least two shooting angle obtain it is more
Open surface state photo, comprising: laptop full face, laptop side photo and laptop turning are shone
Piece.
Each surface state photo is separately input into corresponding detection model by step 102, and according to each detection model
Output is as a result, obtain defects detection result corresponding with each surface state photo.
Wherein, detection model use is in the case where setting shooting angle, the sample image shot for setting intensity of illumination
Training obtains.
The input of detection model is in the case where setting shooting angle, and the surface state shot for setting intensity of illumination is shone
Piece.The output of detection model is the corresponding defects detection result of surface state photo.
Optionally, include: at least one defects detection item in defects detection result, include: defect class in defects detection item
Type, defective locations and defects detection score.
In the present embodiment, based on the sample image under different shooting angles, shot for different intensities of illumination
Different detection models is respectively trained in photo.The sample image photo of different detection models is different.It is input to different inspections
The surface state photo surveyed in model is different.
Optionally, detection model is the model generated based on Fast R-CNN algorithm；In detection model, use is deformable
Convolution constructs convolutional layer, uses focal loss function as loss function, is extracted and inputted using feature pyramid network algorithm
The characteristics of image of photo；And the negative sample image in sample image, it is used in line difficulty sample mining algorithm and excavates to obtain
's.
The present embodiment uses the Fast R-CNN algorithm in object detection algorithms.The theory structure of Fast R-CNN algorithm is such as
Shown in Fig. 1 b.In the case where setting shooting angle, for the sample image that shoots of setting intensity of illumination as the defeated of detection model
Enter, defect type, defective locations and defects detection score are as output.The structure of network is mainly by convolutional layer (conv
Layers), pond layer (Roi pooling), full articulamentum etc. form.Wherein, the convolution operation convolution kernel different using weight
Convolution is scanned to original image or characteristic pattern (feature maps), therefrom extracts the feature of various meanings, and export extremely
In characteristic pattern.Pondization operation then carries out dimensionality reduction operation to characteristic pattern, the main feature in keeping characteristics figure.There is volume using this
Product, the deep neural network model of pondization operation, can be to deformation, the mould for setting the sample image that intensity of illumination is shot
The robustness with higher such as paste, illumination variation, for classification task have it is higher can generalization.
Fast R-CNN algorithm obtains its characteristic pattern first with the convolution operation of disaggregated model.Detection model extracts figure
As the convolution backbone network used when feature is the resnet50 network containing deformable convolution.In traditional convolution, one
Convolution kernel can carry out sampling and convolutional calculation in sliding process, in each position with a fixed size and shape,
Pond layer reduces spatial resolution with fixed ratio, and pond layer is divided into area-of-interest roi fixed characteristic block.Can
It is denaturalized in convolution, has broken this limitation, increase the offset that a convolutional neural networks learn convolution sampled point for convolution kernel
Measure offset, which generates that one wide Gao Yuyuan characteristic pattern is identical, channel be the offset of 2N as illustrated in figure 1 c.So
Design, can reduce sampling of the convolution kernel in training, detection process in noisy background, greatly improve detection model training effect
Rate reduces the false detection rate of qualified products while improving defect recall rate.Fig. 1 c illustrates the standard that convolution kernel size is 3x3 and rolls up
Long-pending and deformable convolution sample mode.It (a) is Standard convolution.(b) (c) (d) is deformable convolution.In normal sample coordinate
Upper to add an offset offset (arrow), wherein (c) special circumstances of (d) as (b), illustrating deformable convolution can be with
As change of scale, the special circumstances of transformation of scale and rotation transformation.
After extracting characteristic pattern, Fast R-CNN algorithm utilizes candidate region network (Region Proposal Network)
Whether include specific object: if carrying out feature comprising object using convolutional network and mentioning if calculating in a certain region of original image
It takes, then detects its object category and boundary box (bounding box)；If not including object, without classification.This
Sample combines the loss of three network branches, does combined training, optimizing detection model parameter.It is defeated when detection model
When the error amount between true value is less than certain threshold value out, deconditioning.
In detection model, uses focal loss function as loss function, it is unbalanced to solve data sample classification
Problem；The characteristics of image of input photo is extracted using feature pyramid network algorithm, improves defect recall rate；Use online difficulty
Sample mining algorithm excavates the negative sample image in sample image, improves detection model precision.
Optionally, each surface state photo is separately input into corresponding detection model, and according to each detection model
Output is as a result, obtain defects detection result corresponding with each surface state photo, comprising: by currently processed surface shape
State photo carries out cutting figure, obtains the local state photo of multiple target sizes, the minimum defect ruler of target size and target product
It is very little to match；Multiple local state photos are separately input into and the currently processed matched detection model of surface state photo
In, obtain defects detection result corresponding with each local state photo；It corresponding with each local state photo will lack
Sunken testing result is combined, and obtains defects detection result corresponding with currently processed surface state photo.
Wherein, a surface state photo is obtained from all surfaces state photo to shine as currently processed surface state
Piece.It carries out currently processed surface state photo to cut figure, obtain and multiple matched mesh of currently processed surface state photo
The local state photo of dimensioning.Target size and the minimum defect size of target product match.That is local state photo
Minimum defect size having a size of target product.Then by with currently processed multiple matched local states of surface state photo
Photo is separately input into and in the currently processed matched detection model of surface state photo, obtains and each local state photo point
Not corresponding defects detection result.Defects detection result corresponding with each local state photo is combined, obtain with
After the corresponding defects detection result of currently processed surface state photo, returns to execution and obtain one from all surfaces state photo
Operation of the surface state photo as currently processed surface state photo is opened, is distinguished until obtaining with each surface state photo
Corresponding defects detection result.
Step 103, basis defects detection corresponding with each surface state photo are as a result, determine the inspection of target product
Survey result.
Optionally, according to defects detection corresponding with each surface state photo as a result, determining the inspection of target product
Survey as a result, may include: according to defects detection corresponding with each surface state photo as a result, and with each detection mould
The corresponding measurement of type refers to weight, determines the testing result of target product.
Measurement is used to measure the reference value of the testing result of detection model output with reference to weight.The inspection of detection model output
The reference value for surveying result is higher, and the corresponding measurement of detection model is higher with reference to weight.
Optionally, previously according to business demand, the corresponding initial measurement of each detection model is set with reference to weight.
In a specific example, according to defects detection corresponding with each surface state photo as a result, and with
The corresponding measurement of each detection model refers to weight, determines that the mode of the testing result of target product can be with are as follows: will with it is each
The corresponding defects detection of surface state photo is as a result, be separately input into decision model, and obtain the detection of decision model output
As a result；Wherein, decision model learns measurement corresponding with each detection model with reference to weight in advance, and measurement is with reference to weight certainly
Iteration updates in the use process of plan model.
In another specific example, according to defects detection corresponding with each surface state photo as a result, and
Measurement corresponding with each detection model refers to weight, determines that the mode of the testing result of target product can be with are as follows: obtain
Defects detection results set corresponding with multiple surface state photos under currently processed shooting angle；In defects detection result
In set, according to the measurement of each defects detection result with reference to the similarity between weight and defects detection result, lacked to each
Sunken testing result is given a mark；According to marking as a result, the matched local testing result of determining and currently processed shooting angle；It will
Local testing result under at least two shooting angle merges processing, obtains the testing result of target product.
Similarity between defects detection result refers to the overlapping degree of each defects detection result.
Optionally, according to defects detection corresponding with each surface state photo as a result, determining the inspection of target product
It surveys as a result, may include: to obtain defects detection knot corresponding with multiple surface state photos under currently processed shooting angle
Fruit set；Each defects detection result in defects detection results set is mapped in same photo；According to each in mapping result
The similarity of defects detection result, the matched local testing result of determining and currently processed shooting angle；At least two are clapped
The part detection taken the photograph under angle, which combines, merges processing, obtains the testing result of target product.
The technical solution of the embodiment of the present application, by obtaining target product to be detected under at least two shooting angle,
Multiple the surface state photos shot at least two intensities of illumination；Then each surface state photo is separately input into
In corresponding detection model, and according to the output of each detection model as a result, obtaining corresponding with each surface state photo
Defects detection result；Detection model use is in the case where setting shooting angle, for the sample image that shoots of setting intensity of illumination
Training obtains；Finally, according to defects detection corresponding with each surface state photo as a result, determining the detection of target product
As a result, can be obtained based on use in the case where setting shooting angle for the sample image training that setting intensity of illumination is shot
Detection model, obtain defects detection corresponding from each surface state photo as a result, can be by different detection models
Testing result, the testing result of a final target product is obtained, so that the testing result of target product is with reference to difference
The testing result of detection model, testing result is more accurate, improves detection accuracy, can be according to the surface state photo of product
Automatic detection is carried out to product, improves detection efficiency.
Second embodiment
Fig. 2 is the flow chart of one of the application second embodiment detection method of product surface state, the present embodiment
Further refinement on the basis of the above embodiments, will be according to defects detection knot corresponding with each surface state photo
Fruit, determine target product testing result refinement are as follows: according to defects detection corresponding with each surface state photo as a result,
And measurement corresponding with each detection model refers to weight, determines the testing result of target product.
It is said below with reference to detection method of the Fig. 2 to a kind of product surface state that the application second embodiment provides
It is bright, comprising the following steps:
Step 201 obtains target product to be detected under at least two shooting angle, at least two intensities of illumination
Shoot multiple obtained surface state photos.
Each surface state photo is separately input into corresponding detection model by step 202, and according to each detection model
Output is as a result, obtain defects detection result corresponding with each surface state photo.
Step 203, according to defects detection corresponding with each surface state photo as a result, and with each detection mould
The corresponding measurement of type refers to weight, determines the testing result of target product.
Wherein, measurement is used to measure the reference value of the testing result of detection model output with reference to weight.Detection model is defeated
The reference value of testing result out is higher, and the corresponding measurement of detection model is higher with reference to weight.
Optionally, previously according to business demand, the corresponding initial measurement of each detection model is set with reference to weight.
Optionally, according to defects detection corresponding with each surface state photo as a result, and with each detection mould
Type it is corresponding measurement refer to weight, determine the testing result of target product, may include: by with each surface state photo pair
The defects detection answered is as a result, be separately input into decision model, and obtain the testing result of decision model output；Wherein, decision
Model learns measurement corresponding with each detection model with reference to weight in advance, and measurement refers to weight using in decision model
Iteration updates in journey.
One decision model of training in advance.The input of decision model is defect corresponding with each surface state photo
Testing result.The output of decision model is the testing result of target product.
It optionally, will be after defects detection result corresponding with each surface state photo be input to decision model, certainly
Plan model is according to defects detection corresponding with each surface state photo as a result, and respectively corresponding with each detection model
Measurement refer to weight, give a mark to each defects detection result, obtain the marking result of each defects detection result.Then it obtains
Marking result is more than that the defects detection result of preset fraction threshold value merges processing, obtains the testing result of target product.
Decision model can be according to the corresponding standard quality inspection of target product as a result, automatic study, updates and different detection moulds
The corresponding measurement of type refers to weight.Standard quality inspection obtains the result is that multiple surface state photos are sent to expert's processing platform
Quality inspection result corresponding with target product.Decision model is according to standard quality inspection as a result, and dividing with each surface state photo
Not corresponding defects detection is as a result, the reference measure weight to each detection model is updated.
Optionally, according to defects detection corresponding with each surface state photo as a result, and with each detection
The corresponding measurement of model is with reference to weight, after the testing result for determining target product, further includes: shine multiple surface state
Piece is sent to expert's processing platform, obtains standard quality inspection result corresponding with target product；By decision model, according to standard matter
Inspection as a result, and defects detection corresponding with each surface state photo as a result, being weighed to the reference measure of each detection model
It is updated again.
Optionally, according to defects detection corresponding with each surface state photo as a result, and with each detection mould
The corresponding measurement of type refers to weight, determines the testing result of target product, may include: to obtain and currently processed shooting
The corresponding defects detection results set of multiple surface state photos under angle；In defects detection results set, according to each
The measurement of defects detection result carries out each defects detection result with reference to the similarity between weight and defects detection result
Marking；According to marking as a result, the matched local testing result of determining and currently processed shooting angle；By at least two shooting angles
Local testing result under degree merges processing, obtains the testing result of target product.
Similarity between defects detection result refers to the overlapping degree of each defects detection result.It is examined for each defect
It surveys as a result, the similarity between defects detection result is higher, marking result is higher.
Each shooting angle corresponds to a defects detection results set.It is corresponding shooting angle in defects detection results set
Multiple surface state photos under degree.A shooting angle is obtained from whole shooting angle as currently processed shooting angle
Degree.Obtain defects detection results set corresponding with multiple surface state photos under currently processed shooting angle.In defect
In testing result set, the similarity between weight and defects detection result is referred to according to the measurement of each defects detection result,
It gives a mark to each defects detection result.According to marking as a result, determining examine with the matched part of currently processed shooting angle
Survey result.For example, obtaining the defects detection result that marking result is more than preset fraction threshold value merges processing, obtain and currently
The matched local testing result of the shooting angle of processing.According to marking as a result, determination is matched with currently processed shooting angle
After local testing result, returns to execution and obtain a shooting angle from whole shooting angle as currently processed shooting angle
Operation, until obtain the local testing result under each shooting angle.
The technical solution of the embodiment of the present application, by according to defects detection knot corresponding with each surface state photo
Fruit, and measurement corresponding with each detection model refer to weight, determine the testing result of target product, can be according to not
With the testing result of detection model and each detection model, corresponding measurement refers to weight, obtains a final mesh
The testing result of product is marked, so that the testing result of target product is referred to reference to the testing result and measurement of different detection models
Weight, testing result is more accurate, improves detection accuracy.
3rd embodiment
Fig. 3 a is the flow chart of one of the application 3rd embodiment detection method of product surface state, the present embodiment
Further refinement on the basis of the above embodiments, will be according to defects detection knot corresponding with each surface state photo
Fruit determines the testing result refinement of target product are as follows: obtain and multiple surface state photos under currently processed shooting angle
Corresponding defects detection results set；Each defects detection result in defects detection results set is mapped in same photo；
It is determining to be tied with the matched part detection of currently processed shooting angle according to the similarity of defects detection result each in mapping result
Fruit；Part detection under at least two shooting angle is combined and merges processing, obtains the testing result of target product.
It is said below with reference to detection method of Fig. 3 a to a kind of product surface state that the application 3rd embodiment provides
It is bright, comprising the following steps:
Step 301 obtains target product to be detected under at least two shooting angle, at least two intensities of illumination
Shoot multiple obtained surface state photos.
Each surface state photo is separately input into corresponding detection model by step 302, and according to each detection model
Output is as a result, obtain defects detection result corresponding with each surface state photo.
Wherein, detection model use is in the case where setting shooting angle, the sample image shot for setting intensity of illumination
Training obtains.
Step 303 obtains a shooting angle as currently processed shooting angle from whole shooting angle.
Wherein, the corresponding defects detection results set of each shooting angle.It is corresponding in defects detection results set
Multiple surface state photos under shooting angle.
Step 304 obtains defects detection knot corresponding with multiple surface state photos under currently processed shooting angle
Fruit set.
Step 305 maps to each defects detection result in defects detection results set in same photo.
Optionally, include: at least one defects detection item in defects detection result, include: defect class in defects detection item
Type, defective locations and defects detection score.According to defective locations, by each defects detection result in defects detection results set
It maps in same photo.
Step 306, according to the similarity of defects detection result each in mapping result, it is determining with currently processed shooting angle
After matched part testing result, returns to execution and obtain a shooting angle from whole shooting angle as currently processed bat
The operation of angle is taken the photograph, until obtaining the local testing result under each shooting angle.
Optionally, it according to the similarity of defects detection result each in mapping result, obtains similarity and is greater than default similarity
The defects detection result of threshold value merges processing, obtains and the matched local testing result of currently processed shooting angle.
Part detection combination under at least two shooting angle is merged processing by step 307, obtains target product
Testing result.
Wherein, the part detection under each shooting angle is combined and merges processing, obtain the detection knot of target product
Fruit.
Fig. 3 b is the integrated stand composition of one of the application 3rd embodiment detection system of product surface state.Pass through
Automated machine equipment shoots to obtain the front of multiple products at least two intensities of illumination in the front of product to be detected
Photo, in the side of product to be detected, for the side photo for multiple products that at least two intensities of illumination are shot, In
The turning of target product to be detected, the turning photos of multiple products shot at least two intensities of illumination to get
To the surface state photo of multiple products.Each surface state photo is carried out cutting figure according to minimum defective proportion, obtain with respectively
The corresponding local state photo of a surface state photo.Local state photo corresponding with each photo is stored to creation data
Library.Multiple local state photos are separately input into corresponding detection model, obtain respectively corresponding with each local state photo
Defects detection result.Defects detection result corresponding with each local state photo is merged, obtains and currently locates
The corresponding defects detection result of the surface state photo of reason.Defects detection result includes: defect classification, defective locations and prediction
Score.Will defects detection corresponding with each surface state photo as a result, being separately input into decision-making module, and obtain decision-making module
The product testing result of output.System run a period of time after, can by Quality Inspector manual review, mark defects detection and
Then the accuracy rate of positioning updates training according to library, to re -training detection model, to be iterated to detection model, to mention
High defects detection accuracy rate.
In a specific example, a kind of detection system of product surface state may include following module: data
Acquisition module, for by mechanical automation equipment, the target product that producing line transport comes successively to be carried out each shooting angle
Image is sent to control module by photograph taking；Fault determination module, for by artificial intelligence vision algorithm, in cluster or
Fault verification is carried out to the image data in the detection request of product surface state on quality inspection all-in-one machine, and returns to testing result；
Control module, for cooperateing with hardware resource behaviour in service, initiation operation, the termination for dispatching Detection task are operated, Detection task
Operation is initiated, multi-model testing result is received, carries out final decision, and issues sub-material instruction；Sub-material module, for by automatic
Change equipment, the in kind of corresponding target product is carried out by sub-material according to the final detection result of decision system；Training engine, is used for base
The image data of Yu expert's mark, carries out deep learning on graphics processor (Graphics Processing Unit, GPU)
Model training；Database, for storing raw image data, model inspection result and the mass analysis data of product.
The photo that data acquisition module on production line generates in real time is converted detection request (query) by control module, and
According to the deployment scenario real-time perfoming load balancing and scheduling of detection model on line, it will test request and be sent to optimal carry
On the server of detection model.On the server real time execution fault determination module.Deep learning in fault determination module
Model is completed via training engine training.Model carries out preset pretreatment for the image data in the monitoring request of arrival
Afterwards, object detection calculating is carried out, and provides the location information for representing the classification information defect of defect, and result is sent to control
Module.Control module designs in conjunction with business scenario, can make and meet to the testing result that model provides according to business demand
The response of production environment scene requirement, such as alarm, storage log, control mechanical arm.Control module can will test result and sound
The processing behavior answered is as production log storage on line into Production database.
In training engine modules, trained model each time can gradually be replaced by the online mode of small flow
The old model just run on line is extended with to achieve the purpose that model with service dynamic extensive.After system runs a period of time,
The accuracy rate of defects detection and positioning can be checked by the information in manually generated database, then update training according to library, weight
New training defects detection model, to improve defects detection accuracy rate.
The technical solution of the embodiment of the present application is shone by obtaining with multiple surface state under currently processed shooting angle
Each defects detection result in defects detection results set is mapped to same photo by the corresponding defects detection results set of piece
In, then according to the similarity of defects detection result each in mapping result, the determining and currently processed matched office of shooting angle
Portion's testing result；Processing is merged finally, the part detection under at least two shooting angle is combined, obtains target product
Testing result can obtain the testing result of a final target product, so that mesh according to the similarity of defects detection result
The testing result of product is marked with reference to the overlapping degree of defects detection result, testing result is more accurate, improves detection accuracy.
Fourth embodiment
Fig. 4 is a kind of structural schematic diagram of the detection device for product surface state that the application fourth embodiment provides, should
The detection device of product surface state, comprising: photo obtains module 401, result obtains module 402 and result determining module
403。
Wherein, photo obtains module 401, for obtaining target product to be detected under at least two shooting angle, needle
Multiple surface state photos that at least two intensities of illumination are shot；As a result module 402 is obtained, is used for each surface state
Photo is separately input into corresponding detection model, and according to the output of each detection model as a result, obtaining and each surface state
The corresponding defects detection result of photo；Wherein, detection model use is in the case where setting shooting angle, for setting intensity of illumination
Obtained sample image training is shot to obtain；As a result determining module 403, for being respectively corresponded according to each surface state photo
Defects detection as a result, determine target product testing result.
The technical solution of the embodiment of the present application, by obtaining target product to be detected under at least two shooting angle,
Multiple the surface state photos shot at least two intensities of illumination；Then each surface state photo is separately input into
In corresponding detection model, and according to the output of each detection model as a result, obtaining corresponding with each surface state photo
Defects detection result；Detection model use is in the case where setting shooting angle, for the sample image that shoots of setting intensity of illumination
Training obtains；Finally, according to defects detection corresponding with each surface state photo as a result, determining the detection of target product
As a result, can be obtained based on use in the case where setting shooting angle for the sample image training that setting intensity of illumination is shot
Detection model, obtain defects detection corresponding from each surface state photo as a result, can be by different detection models
Testing result, the testing result of a final target product is obtained, so that the testing result of target product is with reference to difference
The testing result of detection model, testing result is more accurate, improves detection accuracy, can be according to the surface state photo of product
Automatic detection is carried out to product, improves detection efficiency.
Optionally, as a result determining module 403 includes: the first result determination unit, for shining according to each surface state
The corresponding defects detection of piece as a result, and it is corresponding with each detection model measurement refer to weight, determine target produce
The testing result of product.
Optionally, the first result determination unit includes: that result obtains subelement, and being used for will be corresponding with each surface state photo
Defects detection as a result, be separately input into decision model, and obtain the testing result of decision model output；Wherein, decision model
Type learns measurement corresponding with each detection model with reference to weight in advance, and measurement refers to weight in the use process of decision model
Middle iteration updates.
Optionally, the first result determination unit includes: that set obtains subelement, for obtaining and currently processed shooting angle
The corresponding defects detection results set of multiple surface state photos under degree；As a result it gives a mark subelement, in defects detection knot
In fruit set, according to the measurement of each defects detection result with reference to the similarity between weight and defects detection result, to each
Defects detection result is given a mark；As a result determine subelement, for according to marking as a result, determining with currently processed shooting angle
Matched part testing result；As a result merge subelement, for carrying out the local testing result under at least two shooting angle
Merging treatment obtains the testing result of target product.
Optionally, as a result determining module 403 includes: that set obtains subelement, for obtaining and currently processed shooting angle
The corresponding defects detection results set of multiple surface state photos under degree；As a result subelement is mapped, is used for defects detection knot
Each defects detection result in fruit set maps in same photo；As a result subelement is determined, for according to each in mapping result
The similarity of the defects detection result, the matched local testing result of determining and currently processed shooting angle；As a result merge
Subelement merges processing for combining the part detection under at least two shooting angle, obtains the detection of target product
As a result.
Optionally, as a result obtaining module 402 includes: that photo cuts figure unit, for by currently processed surface state photo
It carries out cutting figure, obtains the local state photo of multiple target sizes, the minimum defect size phase of target size and target product
Match；As a result acquiring unit is matched for being separately input into multiple local state photos with currently processed surface state photo
Detection model in, obtain defects detection result corresponding with each local state photo；As a result combining unit, for will be with
The corresponding defects detection result of each local state photo is combined, and is obtained corresponding with currently processed surface state photo
Defects detection result.
Optionally, detection model is the model generated based on Fast R-CNN algorithm；In detection model, use is deformable
Convolution constructs convolutional layer, uses focal loss function as loss function, is extracted and inputted using feature pyramid network algorithm
The characteristics of image of photo；And the negative sample image in sample image, it is used in line difficulty sample mining algorithm and excavates to obtain
's.
Optionally, target product is laptop；Target product shot under at least two shooting angle obtain it is more
Open surface state photo, comprising: laptop full face, laptop side photo and laptop turning are shone
Piece.
The detection device of product surface state provided by the embodiment of the present application can be performed the application any embodiment and be mentioned
The detection method of the product surface state of confession has the corresponding functional module of execution method and beneficial effect.
5th embodiment
According to an embodiment of the present application, present invention also provides a kind of electronic equipment and a kind of readable storage medium storing program for executing.Fig. 5 is
For realize the embodiment of the present application product surface state detection method electronic equipment block diagram.Electronic equipment is intended to indicate that
Various forms of digital computers, such as, laptop computer, desktop computer, workbench, personal digital assistant, server,
Blade server, mainframe computer and other suitable computer.Electronic equipment also may indicate that various forms of mobile dresses
It sets, such as, personal digital assistant, cellular phone, smart phone, wearable device and other similar computing devices.This paper institute
Component, their connection and the relationship shown and their function are merely exemplary, and are not intended to limit described herein
And/or requirement the application realization.
As shown in figure 5, the electronic equipment includes: one or more processors 501, memory 502, and each for connecting
The interface of component, including high-speed interface and low-speed interface.All parts are interconnected using different buses, and can be pacified
It installs in other ways on public mainboard or as needed.Processor can to the instruction executed in electronic equipment into
Row processing, including storage in memory or on memory (such as, to be coupled to interface in external input/output device
Display equipment) on show GUI graphical information instruction.In other embodiments, if desired, can be by multiple processors
And/or multiple bus is used together with multiple memories with multiple memories.It is also possible to multiple electronic equipments are connected, it is each
Equipment provides the necessary operation in part (for example, as server array, one group of blade server or multiprocessor system
System).In Fig. 5 by taking a processor 501 as an example.
Memory 502 is non-transitory computer-readable storage medium provided herein.Wherein, the memory is deposited
The instruction that can be executed by least one processor is contained, so that at least one described processor executes product provided herein
The detection method of surface state.The non-transitory computer-readable storage medium of the application stores computer instruction, which refers to
It enables for making computer execute the detection method of product surface state provided herein.
Memory 502 is used as a kind of non-transitory computer-readable storage medium, can be used for storing non-instantaneous software program, non-
Instantaneous computer executable program and module, such as the corresponding journey of detection method of the product surface state in the embodiment of the present application
Sequence instruction/module is (for example, attached photo shown in Fig. 4 obtains module 401, result obtains module 402 and result determining module
403).Non-instantaneous software program, instruction and the module that processor 501 is stored in memory 502 by operation, thereby executing
The various function application and data processing of server, i.e. the detection side of product surface state in realization above method embodiment
Method.
Memory 502 may include storing program area and storage data area, wherein storing program area can store operation system
Application program required for system, at least one function；Storage data area can be stored according to the detection side for realizing product surface state
The electronic equipment of method uses created data etc..In addition, memory 502 may include high-speed random access memory, also
It may include non-transitory memory, a for example, at least disk memory, flush memory device or other non-instantaneous solid-state memories
Part.In some embodiments, it includes the memory remotely located relative to processor 501 that memory 502 is optional, these are remotely deposited
Reservoir can be by being connected to the network to the electronic equipment for the detection method for realizing product surface state.The example of above-mentioned network includes
But be not limited to internet, intranet, local area network, mobile radio communication and combinations thereof.
The electronic equipment of the detection method of product surface state can also include: input unit 503 and output device 504.
Processor 501, memory 502, input unit 503 and output device 504 can be connected by bus or other modes, Fig. 5
In by by bus connect for.
Input unit 503 can receive the number or character information of input, and generate the inspection with realization product surface state
The related key signals input of the user setting and function control of the electronic equipment of survey method, such as touch screen, keypad, mouse
The input units such as mark, track pad, touch tablet, indicating arm, one or more mouse button, trace ball, control stick.Output device
504 may include display equipment, auxiliary lighting apparatus (for example, LED) and haptic feedback devices (for example, vibrating motor) etc..It should
Display equipment can include but is not limited to, and liquid crystal display (LCD), light emitting diode (LED) display and plasma are shown
Device.In some embodiments, display equipment can be touch screen.
The various embodiments of system and technology described herein can be in digital electronic circuitry, integrated circuit system
It is realized in system, dedicated ASIC (specific integrated circuit), computer hardware, firmware, software, and/or their combination.These are various
Embodiment may include: to implement in one or more computer program, which can be
It executes and/or explains in programmable system containing at least one programmable processor, which can be dedicated
Or general purpose programmable processors, number can be received from storage system, at least one input unit and at least one output device
According to and instruction, and data and instruction is transmitted to the storage system, at least one input unit and this at least one output
Device.
These calculation procedures (also referred to as program, software, software application or code) include the machine of programmable processor
Instruction, and can use programming language, and/or the compilation/machine language of level process and/or object-oriented to implement these
Calculation procedure.As used herein, term " machine readable media " and " computer-readable medium " are referred to for referring to machine
It enables and/or data is supplied to any computer program product, equipment, and/or the device of programmable processor (for example, disk, light
Disk, memory, programmable logic device (PLD)), including, receive the machine readable of the machine instruction as machine-readable signal
Medium.Term " machine-readable signal " is referred to for machine instruction and/or data to be supplied to any of programmable processor
Signal.
In order to provide the interaction with user, system and technology described herein, the computer can be implemented on computers
The display device for showing information to user is included (for example, CRT (cathode-ray tube) or LCD (liquid crystal display) monitoring
Device)；And keyboard and indicator device (for example, mouse or trace ball), user can by the keyboard and the indicator device come
Provide input to computer.The device of other types can be also used for providing the interaction with user；For example, being supplied to user's
Feedback may be any type of sensory feedback (for example, visual feedback, audio feedback or touch feedback)；And it can use
Any form (including vocal input, voice input or tactile input) receives input from the user.
System described herein and technology can be implemented including the computing system of background component (for example, as data
Server) or the computing system (for example, application server) including middleware component or the calculating including front end component
System is (for example, the subscriber computer with graphic user interface or web browser, user can pass through graphical user circle
Face or the web browser to interact with the embodiment of system described herein and technology) or including this backstage portion
In any combination of computing system of part, middleware component or front end component.Any form or the number of medium can be passed through
Digital data communicates (for example, communication network) and is connected with each other the component of system.The example of communication network includes: local area network
(LAN), wide area network (WAN) and internet.
Computer system may include client and server.Client and server is generally off-site from each other and usually logical
Communication network is crossed to interact.By being run on corresponding computer and each other with the meter of client-server relation
Calculation machine program generates the relationship of client and server.
The technical solution of the embodiment of the present application, by obtaining target product to be detected under at least two shooting angle,
Multiple the surface state photos shot at least two intensities of illumination；Then each surface state photo is separately input into
In corresponding detection model, and according to the output of each detection model as a result, obtaining corresponding with each surface state photo
Defects detection result；Detection model use is in the case where setting shooting angle, for the sample image that shoots of setting intensity of illumination
Training obtains；Finally, according to defects detection corresponding with each surface state photo as a result, determining the detection of target product
As a result, can be obtained based on use in the case where setting shooting angle for the sample image training that setting intensity of illumination is shot
Detection model, obtain defects detection corresponding from each surface state photo as a result, can be by different detection models
Testing result, the testing result of a final target product is obtained, so that the testing result of target product is with reference to difference
The testing result of detection model, testing result is more accurate, improves detection accuracy, can be according to the surface state photo of product
Automatic detection is carried out to product, improves detection efficiency.
It should be understood that various forms of processes illustrated above can be used, rearrangement increases or deletes step.Example
Such as, each step as described in this application can be performed in parallel or be sequentially performed the order that can also be different and execute, only
It is desired as a result, being not limited herein to can be realized technical solution disclosed in the present application.
Above-mentioned specific embodiment does not constitute the limitation to the application protection scope.Those skilled in the art should be bright
White, according to design requirement and other factors, various modifications can be carried out, combination, sub-portfolio and substitution.It is any in the application
Spirit and principle within made modifications, equivalent substitutions and improvements etc., should be included within the application protection scope.
Claims (11)
1. a kind of detection method of product surface state characterized by comprising
Target product to be detected is obtained under at least two shooting angle, is shot at least two intensities of illumination more
Open surface state photo；
Each surface state photo is separately input into corresponding detection model, and according to the output of each detection model
As a result, obtaining defects detection result corresponding with each surface state photo；
Wherein, the detection model use is in the case where setting shooting angle, the sample image shot for setting intensity of illumination
Training obtains；
According to defects detection corresponding with each surface state photo as a result, determining the detection knot of the target product
Fruit.
2. the method according to claim 1, wherein according to corresponding with each surface state photo
Defects detection is as a result, determine the testing result of the target product, comprising:
According to defects detection corresponding with each surface state photo as a result, and with each detection model point
Not corresponding measurement refers to weight, determines the testing result of the target product.
3. according to the method described in claim 2, it is characterized in that, according to corresponding with each surface state photo
Defects detection as a result, and measurement corresponding with each detection model refer to weight, determine the target product
Testing result, comprising:
Will defects detection corresponding with each surface state photo as a result, be separately input into decision model, and described in obtaining
The testing result of decision model output；
Wherein, the decision model learns measure with reference to weight, the measurement corresponding with each detection model in advance
With reference to weight, iteration updates in the use process of the decision model.
4. according to the method described in claim 2, it is characterized in that, according to corresponding with each surface state photo
Defects detection as a result, and measurement corresponding with each detection model refer to weight, determine the target product
Testing result, comprising:
Obtain defects detection results set corresponding with multiple surface state photos under currently processed shooting angle；
In the defects detection results set, weight and defects detection are referred to according to the measurement of each defects detection result
As a result the similarity between gives a mark to each defects detection result；
According to marking as a result, the matched local testing result of the determining and currently processed shooting angle；
Local testing result under at least two shooting angle is merged into processing, obtains the detection of the target product
As a result.
5. the method according to claim 1, wherein according to corresponding with each surface state photo
Defects detection is as a result, determine the testing result of the target product, comprising:
Obtain defects detection results set corresponding with multiple surface state photos under currently processed shooting angle；
Each defects detection result in the defects detection results set is mapped in same photo；
According to the similarity of the defects detection result each in the mapping result, the determining and currently processed shooting angle
Matched part testing result；
Part detection under at least two shooting angle is combined and merges processing, obtains the detection of the target product
As a result.
6. method according to claim 1-5, which is characterized in that each surface state photo difference is defeated
Enter into corresponding detection model, and according to the output of each detection model as a result, obtaining shining with each surface state
The corresponding defects detection result of piece, comprising:
It carries out currently processed surface state photo to cut figure, obtains the local state photo of multiple target sizes, the target
Size and the minimum defect size of the target product match；
Multiple described local state photos are separately input into and the matched detection mould of currently processed surface state photo
In type, defects detection result corresponding with each local state photo is obtained；
Defects detection result corresponding with each local state photo is combined, obtain with it is described currently processed
The corresponding defects detection result of surface state photo.
7. method according to claim 1-5, it is characterised in that:
The detection model is the model generated based on FastR-CNN algorithm；
In the detection model, convolutional layer is constructed using deformable convolution, uses focalloss function as loss function,
The characteristics of image of input photo is extracted using feature pyramid network algorithm；And
Negative sample image in the sample image is used in what line difficulty sample mining algorithm excavated.
8. method according to claim 1-5, which is characterized in that the target product is laptop；
Multiple surface state photos that the target product is shot under at least two shooting angle, comprising: notebook electricity
Brain full face, laptop side photo and laptop turning photo.
9. a kind of detection device of product surface state characterized by comprising
Photo obtains module, for obtaining target product to be detected under at least two shooting angle, at least two light
Take multiple surface state pictures that intensity is shot；
As a result module is obtained, for each surface state photo to be separately input into corresponding detection model, and according to each
The output of the detection model is as a result, obtain defects detection result corresponding with each surface state photo；
Wherein, the detection model use is in the case where setting shooting angle, the sample image shot for setting intensity of illumination
Training obtains；
As a result determining module, for basis with each corresponding defects detection of surface state photo as a result, determining institute
State the testing result of target product.
10. a kind of electronic equipment characterized by comprising
At least one processor；And
The memory being connect at least one described processor communication；Wherein,
The memory is stored with the instruction that can be executed by least one described processor, and described instruction is by described at least one
It manages device to execute, so that at least one described processor is able to carry out method of any of claims 1-8.
11. a kind of non-transitory computer-readable storage medium for being stored with computer instruction, which is characterized in that the computer refers to
It enables for making the computer perform claim require method described in any one of 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910816859.2A CN110517259A (en) | 2019-08-30 | 2019-08-30 | A kind of detection method, device, equipment and the medium of product surface state |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910816859.2A CN110517259A (en) | 2019-08-30 | 2019-08-30 | A kind of detection method, device, equipment and the medium of product surface state |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110517259A true CN110517259A (en) | 2019-11-29 |
Family
ID=68628872
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910816859.2A Pending CN110517259A (en) | 2019-08-30 | 2019-08-30 | A kind of detection method, device, equipment and the medium of product surface state |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110517259A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110261392A (en) * | 2019-06-19 | 2019-09-20 | 北京百度网讯科技有限公司 | Quality determining method, device, electronic equipment and system |
| CN111325744A (en) * | 2020-03-06 | 2020-06-23 | 成都数之联科技有限公司 | Sample processing method in panel defect detection |
| CN111402220A (en) * | 2020-03-11 | 2020-07-10 | 北京百度网讯科技有限公司 | Method and device for acquiring information |
| CN111768386A (en) * | 2020-06-30 | 2020-10-13 | 北京百度网讯科技有限公司 | Product defect detection method and device, electronic equipment and storage medium |
| CN112184622A (en) * | 2020-08-31 | 2021-01-05 | 广东三和管桩股份有限公司 | Pipe pile appearance quality inspection method and system, computer equipment and storage medium |
| CN112686852A (en) * | 2020-12-25 | 2021-04-20 | 浙江伟星实业发展股份有限公司 | Product defect identification method, device, equipment and storage medium |
| CN112907560A (en) * | 2021-03-16 | 2021-06-04 | 中科海拓(无锡)科技有限公司 | Notebook appearance flaw segmentation method based on deep learning |
| CN113176278A (en) * | 2021-03-19 | 2021-07-27 | 哈工大机器人(中山)无人装备与人工智能研究院 | Panel defect detection equipment and panel defect detection method |
| CN113192056A (en) * | 2021-05-21 | 2021-07-30 | 北京市商汤科技开发有限公司 | Image detection method and related device, equipment and storage medium |
| CN113252678A (en) * | 2021-03-24 | 2021-08-13 | 上海万物新生环保科技集团有限公司 | Appearance quality inspection method and equipment for mobile terminal |
-
2019
- 2019-08-30 CN CN201910816859.2A patent/CN110517259A/en active Pending
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110261392A (en) * | 2019-06-19 | 2019-09-20 | 北京百度网讯科技有限公司 | Quality determining method, device, electronic equipment and system |
| CN111325744A (en) * | 2020-03-06 | 2020-06-23 | 成都数之联科技有限公司 | Sample processing method in panel defect detection |
| WO2021179565A1 (en) * | 2020-03-11 | 2021-09-16 | 北京百度网讯科技有限公司 | Method and apparatus for acquiring information |
| CN111402220A (en) * | 2020-03-11 | 2020-07-10 | 北京百度网讯科技有限公司 | Method and device for acquiring information |
| CN111402220B (en) * | 2020-03-11 | 2023-06-09 | 北京百度网讯科技有限公司 | Method and device for acquiring information |
| CN111768386A (en) * | 2020-06-30 | 2020-10-13 | 北京百度网讯科技有限公司 | Product defect detection method and device, electronic equipment and storage medium |
| CN111768386B (en) * | 2020-06-30 | 2024-02-20 | 北京百度网讯科技有限公司 | Product defect detection method, device, electronic equipment and storage medium |
| CN112184622A (en) * | 2020-08-31 | 2021-01-05 | 广东三和管桩股份有限公司 | Pipe pile appearance quality inspection method and system, computer equipment and storage medium |
| CN112686852A (en) * | 2020-12-25 | 2021-04-20 | 浙江伟星实业发展股份有限公司 | Product defect identification method, device, equipment and storage medium |
| CN112907560A (en) * | 2021-03-16 | 2021-06-04 | 中科海拓(无锡)科技有限公司 | Notebook appearance flaw segmentation method based on deep learning |
| CN113176278B (en) * | 2021-03-19 | 2022-07-22 | 哈工大机器人(中山)无人装备与人工智能研究院 | Panel defect detection equipment and panel defect detection method |
| CN113176278A (en) * | 2021-03-19 | 2021-07-27 | 哈工大机器人(中山)无人装备与人工智能研究院 | Panel defect detection equipment and panel defect detection method |
| CN113252678A (en) * | 2021-03-24 | 2021-08-13 | 上海万物新生环保科技集团有限公司 | Appearance quality inspection method and equipment for mobile terminal |
| CN113192056A (en) * | 2021-05-21 | 2021-07-30 | 北京市商汤科技开发有限公司 | Image detection method and related device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110517259A (en) | A kind of detection method, device, equipment and the medium of product surface state | |
| US20220036533A1 (en) | Image defect detection method and apparatus, electronic device, storage medium and product | |
| CN109949286A (en) | Method and apparatus for output information | |
| CN110070536A (en) | A kind of pcb board component detection method based on deep learning | |
| CN109993734A (en) | Method and apparatus for output information | |
| CN109584227A (en) | A kind of quality of welding spot detection method and its realization system based on deep learning algorithm of target detection | |
| CN110009614A (en) | Method and apparatus for output information | |
| CN108257121B (en) | Method, apparatus, storage medium and the terminal device that product defects detection model updates | |
| CN106875373A (en) | Mobile phone screen MURA defect inspection methods based on convolutional neural networks pruning algorithms | |
| CN107064170A (en) | One kind detection phone housing profile tolerance defect method | |
| CN108830837A (en) | A kind of method and apparatus for detecting ladle corrosion defect | |
| CN109544548A (en) | Defect inspection method, device, server, equipment and the storage medium of cutlery box | |
| CN109978870A (en) | Method and apparatus for output information | |
| CN109858414A (en) | A kind of invoice piecemeal detection method | |
| CN111881908B (en) | Target detection model correction method, detection device, equipment and medium | |
| CN110197004B (en) | Circuit simulation method and device based on mobile terminal, computer medium and equipment | |
| WO2019108371A1 (en) | Training neural networks to detect similar three-dimensional objects using fuzzy identification | |
| JP2021515885A (en) | Methods, devices, systems and programs for setting lighting conditions and storage media | |
| CN110458791A (en) | Mass defect detection method and detection device | |
| CN113537496A (en) | Deep learning model visual construction system and application and design method thereof | |
| EP3882817A2 (en) | Method, apparatus and device for recognizing bill and storage medium | |
| CN111353603A (en) | Deep learning model individual prediction interpretation method | |
| CN114996310A (en) | Data processing method, device, server and storage medium | |
| CN113592017A (en) | Deep learning model standardization training method, management system and processing terminal | |
| CN116068057B (en) | SOT packaging chip defect detection method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |